Recently, magnetic resonance imaging has changed the way the world looks at radiology. With the evolution of magnetic resonance imaging, quality assurance images of the biological structure of a patient can now be produced. However, conventional MRI reconstruction methods often fail. This is because they are affected by poor image quality due to artifacts and noise, not to mention the long imaging times. Such issues have caused many to turn their heads towards GANs. • Magnetic Resonance Imaging (MRI) is a critical tool in medical diagnostics, yet its acquisition speed remains a challenge. This report presents the challenges of Client Selection(CS), cGAN’s with Federated Learning (FL) to enhance MRI reconstruction from under-sampled k-space data; Leading to faster convergence, improved model accuracy, and more efficient use of resources • Using the privacy-preserving benefits of FL with (CS) and the generative capabilities of cGANs, the proposed model achieves the promising approaches of federated learning by addressing the concerns that enable collaborative model training across multiple decentralized institutions without sharing sensitive patient data. Reconstructing MRI using federated learning aims to achieve both high-quality and accurate image reconstruction while maintaining data confidentiality. • In this research paper, we aim to achieve high reconstruction accuracy and data confidentiality by proposing a new approach for reconstructing MRI using federated learning. • First, this dissertation proposes a framework for adaptive client selection for federated MRI reconstruction that combines Dual Conditional GANs (Dual-cGANs) and federated learning (FL) to respond to important issues in the medical imaging landscape: Privacy-preservation: Enables collaborative efforts across multiple institutions without sharing the raw data. Heterogeneous Data that leverages a client-local model to accommodate non-IID data. Resource Efficiency achieves 95/100 accuracy (compared to 90/100 for FedAVG) while reducing communication costs by 45/100 and converging 40/100 faster. Reconstruction Quality which produces realistic MRI images conditioned on the medical attributes of patients (i.e., age, sex, health condition) using Dual-cGANs. Key original contributions include: Client scoring (optimally weighted: Si = α ·Mi + β · Qi + γ · Pi The ability to leverage FL-GAN dynamics to create synthetic data for domain adaptation. Resource-aware selection: Selected clients with substantial contributions to federated training and reduced the effect of stragglers. Evaluated against the IXI and FastMRI datasets, the approach is shown to improve upon accuracy and privacy for practitioners while maintaining the ability to scale for clinical use. • Finally, experimental results were conducted via Google Colaboratory GPU Environment, which is a cloud-based platform that typically provides access to NVIDIA Tesla series GPUs to demonstrate significant improvements in reconstruction fidelity, highlighting superior accuracy, privacy, and scalability for clinical deployment.